Patch to check for Integer overflow in presence of casts.

[Richard Smith]

On Tue, Oct 14, 2014 at 1:29 PM, Reid Kleckner <rnk at google.com> wrote:

> On Tue, Oct 14, 2014 at 11:38 AM, jahanian <fjahanian at apple.com> wrote:
>
>> Currently we do not warn Integer overflow in presence of casts; as in:
>> unsigned long long l = 65536 * 65536;
>>
>> This small patch fixes it:
>>
>
> lgtm, thanks!
>
> Also, is there a reason we do not warn on overflow for unsigned; as in:
>> unsigned long long l2 = 65536 * (unsigned)65536;
>> Even though resulting constant is the unexpected 0.
>>
>
> The result is defined to be zero. We could warn, but false positives are
> likely.
>

If the thing that happens *immediately* after an unsigned overflow is an
implicit cast to a larger type (in which the computed value would have been
different), that might suggest that the overflow was unintended. Might be
worth investigating how large the false positive rate is for that case.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20141014/62f1155d/attachment.html>